DE102011004160A1 - Method and device for examining a hollow organ with a magnet-guided endoscope capsule - Google Patents

Abstract

In a method and a device for examining a hollow organ (6) with a magnet-guided endoscope capsule (2), with which video images are continuously generated from the interior of the hollow organ (6), these are recorded during the navigation of the endoscope capsule (2) in the hollow organ (6 ) evaluated with methods of digital image processing and automatically analyzed for the presence of diagnostically relevant information, and the result of this analysis is used to control the endoscope capsule (2).

Description

The invention relates to a method and a device for examining a hollow organ with a magnet-guided endoscope capsule.

For the optical examination of the inner surface of a living body in the body of a hollow organ, for example, the gastrointestinal tract of a patient, it is for example from the DE 101 42 253 C1 known to introduce into the hollow organ an endoscope capsule in the study of the gastrointestinal tract, for example, by swallowing, which can be freely maneuvered without solid compound to the outside in the hollow organ by external magnetic fields.

In such an endoscopic examination (magnetically guided capsule endoscopy), the physician carrying out the diagnosis must process a large number of images consecutive at short intervals on the one hand with regard to possibly existing diagnostically relevant image contents, on the other hand also with regard to a correct control of the endoscope capsule. However, the control of the endoscope capsule requires a high degree of attention, so that in principle there is a risk that the examining doctor overlooks diagnostically relevant details.

Although it is generally known in passive capsule endoscopy to subsequently subject the existing image material to digital image processing in order to provide methods of computer-aided diagnosis (CAD) methods with which it is possible to automatically acquire clinically relevant pathological structures recognize and thus facilitate the evaluation of the examining doctor.

However, such a subsequent automated evaluation would also take place in the magnet-guided capsule endoscopy on the basis of the available video material, so that there is still the risk that pathological manifestations are not or only indistinctly reproduced on the video images, as these from the endoscope capsule or only have been detected indistinctly because the examining doctor has not recognized the diagnostic relevance during the study and has not taken into account in the control of the magnetically guided endoscope capsule.

The invention is therefore based on the object of specifying a method for examining a hollow organ with a magnet-guided endoscope capsule, which increases the probability with which pathological structures in the hollow organ can be detected, and accordingly reduces the probability of false-negative or false-positive diagnosis is. In addition, the invention has the object to provide a device operated by this method.

The above object is achieved according to the invention, with a method having the features of claim 1. According to these features, the video images generated from the interior of the hollow organ continuously with the magnet-guided endoscope capsule are evaluated during navigation of the endoscope capsule in the cavity with methods of digital image processing and automatically analyzed for the presence of diagnostically relevant information, and the result of the analysis, i. H. a statement about whether the video images are a diagnostically relevant, d. H. contain information indicative of a disease or not, used to control the endoscope capsule.

The indicative of a disease, d. H. Diagnostically relevant information can be derived, for example, directly from a recognizable in the video image pathological change on the wall of the hollow organ or indirectly from the Opacityseigenschaften the fluid contained in the hollow organ, as certain clinical pictures also by turbidity or coloration of the liquid or by the presence of particles in the Make fluid noticeable. Of particular interest is the detection of acute gastritis (gastritis), as this is a major cause of gastric cancer. Preferably, therefore, the diagnostic relevant information is the presence of areas on the wall of the hollow organ which indicate possible gastritis.

The invention is based on the consideration that the recognition of diagnostically relevant areas of the wall or diagnostically relevant properties of the liquid in which the endoscope capsule floats can be used during the examination to control the endoscope capsule in such a way as to identify the possibly pathologically identified wall area subjecting the hollow organ or the irregularities discovered in the fluid itself to a more extensive endoscopic examination, for example by stopping the endoscope capsule and moving it closer to the relevant surface area in order to allow the physician a closer examination. This may be, for example, a higher magnification or a viewing from different angles, which allows it to better judge whether it is indeed a gastric region or not. Upon detection of a diagnostically relevant change in the fluid, for example, the endoscope capsule can be rotated in order to determine the properties of the fluid in analyze different directions in more detail, for example, to be able to locate the cause of the change, such as the location on the wall, where a detachment of particles takes place, so that it can be examined in more detail.

This control can either be automated or performed manually by the doctor if visually or visually such an indication of the possible presence of a pathological change is displayed when viewing the video images on the monitor.

For this purpose, in a preferred embodiment of the method, a grayscale image is generated from the video images for locating high intensity gradients by preprocessing, whose pixels are each assigned a gray value from which a filtered image is generated by smoothing and edge filtering, whose pixels are assigned a filter value, wherein a threshold value is derived from the filter values of the pixels of the filtered image. The video image is divided into a plurality of groups, each comprising a multiplicity of pixels, and only those groups which contain a minimum number of pixels which exceed the threshold value in the filtered image are used as relevant groups for analysis. These measures reduce the number of pixels needed for analysis to a level that allows for automated on-the-fly analysis.

A particularly high accuracy is achieved if each relevant group is assigned a multitude of properties, with the aid of which an expert system trained on the basis of evaluated test images is used to assess whether the relevant group indicates a gastric area or not.

With regard to the device, the object is achieved according to the invention with a device having the features of claim. 7

For further explanation of the invention reference is made to the figures. Show it:

1 a device according to the invention in a schematic schematic diagram,

2a , b a flowchart in which the sequence of the automatic analysis of the video images provided by the endoscope capsule is represented,

3 and 4 also schematically simplified images of the interior of a stomach with different pathological manifestations within a liquid in the stomach.

According to 1 there is a magnetic guided endoscope capsule 2 in one with a liquid 4 filled hollow organ 6 for example, a stomach. With at least one in the endoscope capsule 2 arranged video camera 8th Video images are continuously recorded and transmitted via radio to an external control and evaluation device 10 transmitted. In the endoscope capsule 2 is a permanent magnet 12 arranged. The guidance, ie the movement and orientation of the endoscope capsule 2 takes place with the help of an inhomogeneous magnetic field H, that of an external magnet system 14 is generated, for example, comprises three magnetic coils whose coil axes are oriented perpendicular to each other and define a device reference system xyz.

In the example shown is in the field of view 16 the video camera on the wall 18 of the hollow organ a pathological structure 20 , for example, a local inflammation (gastric area). The from the video camera 8th recorded and sent to the control and evaluation 10 transmitted video images are displayed on a monitor 22 played. The pathological structure 20 is on the monitor 22 to recognize and highlighted by hatching.

The magnet system 14 is controlled by means of control signals S from the control and evaluation 10 either automatically or by inputs from a user via an external input device 24 , in the example illustrated by a joystick, are generated.

In the control and evaluation unit 10 are implemented image processing algorithms with which it is possible pathological structures 20 automatically detect and mark in the video image, as indicated by the hatching. In addition, other visual or acoustic displays can be found in the 1 through a speaker 26 symbolizes, provided that the operator to the presence of one of the control and evaluation 10 as a pathologically classified structure 20 to give them the possibility of this possibly pathological structure 20 to investigate, for example, the endoscope capsule 2 stops or closer to the structure 20 approaches in order to be able to judge with certainty whether it is really a pathological phenomenon or not.

The in the control and evaluation unit 10 image processing and image evaluation is in 2a and b is illustrated with reference to a flowchart. In a first step 30 the original color video image O = v (x, y) is stored, each pixel x, y being assigned a three-channel RGB value v. For further processing will now be in a next step 32 from this video picture O = v (x, y) a gray value image G = g (x, y) is calculated, wherein each pixel x, y is assigned a gray value g on a scale comprising, for example, 256 gray scale values. The basis for this is the consideration that there are pronounced intensity changes between gastritic areas and non-gastritic areas, which are also found in the single-channel gray scale image G. In this gray-scale image G, high intensity gradients can then be localized with 2D filters, for example after previous smoothing with an edge filter. The gray value image G is accordingly in step 34 is convoluted with a filter kernel K, which is preferably itself a convolution of a Gaussian filter and a Laplace filter (a so-called L (aplacian) o (f) G (Gaussian) filter) to regions with pronounced edges where the Intensity with high spatial frequency changes to highlight. This results in a filtered image F = G × K = f (x, y). The number of columns or rows of the filter core used here corresponds approximately to the number of pixels (pixels) which are located in a transition zone forming the edge between an inflamed area and a healthy area. In practice, a filter with 9 × 9 = 81 matrix elements with a variance σ = 1.4 has proven itself.

In this filtered image F = f (x, y) is now in step 36 searched for a first threshold t _F , with the help of which, relevant amplitude changes of non-relevant amplitude changes can be separated. Since the video images produced differ considerably in terms of their contrast and the lighting conditions, a fixed threshold value is unsuitable for this purpose. For the determination of the first threshold value t _F , therefore, for each filtered image F = f (x, y), a histogram with, for example, 256 equal intervals is formed whose lower limit is formed by the minimum intensity value and whose upper limit by the maximum intensity value. For this purpose, the histogram is used to analyze at which point one quarter of the maximum intensity value is undershot. With the threshold value t _F set in this manner, a mask or a binary image is formed which assumes the value 1 for those pixels whose intensity is greater than the threshold value and assumes the value 0 for the remaining pixels.

In a next step 38 Now, the filtered image F = f (x, y) is divided into a plurality i = 1, ..., n of disjoint groups (Bins) bin _i (x, y), each comprising a plurality of pixels x, y , where the set of groups bin _i (x, y) covers the whole picture. In the example, groups bin _i (x, y) are formed from 15 × 15 pixels each. These groups _i (x, y) are now in a next step 40 is analyzed as to how many of the pixels x, y within such a group bin _i (x, y) have a (filtered) gray value f that exceeds a second predetermined threshold value t _s , ie the number N _{i is} within the group bin _i (x, y) existing pixels x, y for which the relationship g (x, y)> t _{s is} satisfied. For further processing, only those groups bin _i (x, y) are taken into account, in which the number N _{i of} such pixels is greater than a minimum value N ₀ , which is set at 10 in the example. In other words, at least 10 of the 225 pixels x, y of a group bin _i (x, y) must have a (filtered) gray value f that exceeds the threshold value t _s . This approach is based on the consideration that both fast intensity changes in the image, such as occur in a gastric lesion, and high-frequency noise lead to an accumulation of filter responses with large amplitude changes, the latter occurring only in isolation and with a small number. In this way, remain in step 42 relevant groups are _{i, rel} (x, y), which in the original image mark those areas that are subjected to further analysis.

In a next step 44 the original image data v (x, y) of the video image O of these relevant groups bin _{i, rel} (x, y), ie the original RGB values of the individual remaining relevant pixels x, y, are used. The original image data v of each relevant group bin _{i, rel} (x, y) will now be in step 46 analyzed for a variety of properties. These are in the present case color properties, properties of the color histogram, color structure properties, texture properties and filter properties. An example of color properties is the mean and the maximum value for each channel of the RGB color space within such a group is _{i i, rel} (x, y). In addition, color histograms for the R channel in the RGB color space, the third channel in the HSV color space and the L channel of the Lab color space are formed.

The color structure properties serve to distinguish between the quantitative and spatial distribution of color properties within a group bin _{i, rel} (x, y). For this purpose, the groups bin _{i, rel} (x, y) are subdivided into subgroups. In the exemplary embodiment, 25 subgroups each of 3 × 3 pixels are formed in each of the groups bin _{i, rel} (x, y). Now in five color channels (RGB color space: R, G, B, Lab color space: L, HSV color space: V) the occurrence of a special color property is detected. As a feature, the frequency of occurrence of a particular color characteristic in one of these channels within the subgroup is then recorded.

For examining the texture properties within each relevant group _{i, rel} (x, y) is also related to the gray values g of the unfiltered gray value image G present in this group and used a so-called difference-of-boxes filter whose filter responses have high negative and positive amplitudes for spatial changes in intensity.

In this way, each relevant group bin _{i, rel} (x, y) is assigned a property vector e with a number of components corresponding to the number m of properties.

In previous learning steps, for a plurality of test images, the original image data of the relevant groups bin _{i, rel} (x, y) extracted in each of the test images according to the method explained above are evaluated by an expert for their diagnostic relevance. For this purpose, two classes C1, C2 are formed. Relevant groups bin _{i, rel} (x, y) with positive findings (diagnostically relevant) are assigned to class C1, the remaining (diagnostically irrelevant) to class C2 (step 48 ).

With the aid of a learning algorithm, in the example a so-called adaptive boosting algorithm, the system is trained in such a way on the basis of the characteristic vector bin _{i, rel} (x, y) associated with each relevant group bin _{i, rel} (x, y) high probability can decide which class C1, C2 this group _{i, rel} (x, y) can be assigned.

In the course of this training phase, the properties used for the later automated image analysis, ie the actually necessary and suitable, ie relevant components of the property vector, are also extracted from a larger number of properties by disregarding those components that are used in the training phase for both Groups are _{i, rel} (x, y) of class C1 as well as the groups bin _{i, rel} (x, y) of class C2 the value 0. Equally, those features whose variance in group C1 is greater than or nearly equal to the variance of that feature in the total number of groups are still considered.

These measures have resulted in practice in a number of 66 features that are used in the automated image analysis to assess whether it is a gastric lesion or not.

The expert system trained in this way will now be in steps 46 and 48 judges which class C1, C2 the video image produced in an endoscopic examination with the steps 30 to 42 identified relevant groups bin _{i, rel are} assigned. In other words, the expert system automatically judges whether or not the relevant group bin _{i rel} indicates a gastric region. If relevant groups bin _{i, rel of} a video image of class C1 have been assigned, this is indicated acoustically or optically, for example within the video image, by a corresponding highlighting to the user, so that he can actively intervene in the control of the endoscope capsule. In addition or as an alternative to such a manual control, automatic control of the endoscope capsule may also be provided.

The method used for the detection of gastritis is basically suitable with suitable preprocessing steps and suitable for the characterization of property vectors for the automatic detection of other diseases in which the diagnostically relevant information in a turbidity and / or coloration of a fluid located in the hollow organ and / or a presence of particles in the liquid, d. H. in which the optical properties of the fluid indicate the presence of a clinical picture.

Thus, in certain diseases in the examined organ mixes the supplied liquid whose properties are known with foreign bodies. For example, these foreign bodies can be tissue parts which detach from the stomach wall in the manner of a flake in a gastric ulcer. These tissue flakes are then recognized in the endoscope image as floating in the liquid particles. The type, amount and size of these foreign bodies provide a very good first indication of whether a particular disease is present, even if the affected anatomical site has not yet been sighted with the endoscope. To recognize the foreign body in the endoscope usually as a floating particles, but may also be formed in small size as clouding of the liquid.

In 3 is schematically reproduced an endoscopic image in which as a result of a gastric ulcer exudate has detached from the stomach wall and as whitish fluffy foreign body 50 floating in the liquid. In an automatic detection can be exploited that these flakes 50 opposite to those in the background located on the stomach wall, as well as the stomach wall reddish structures 52 be clearly visible as bright white image areas and thus automatically segmented and analyzed in terms of size and number.

In 4 another clinical picture is presented in which bile is a yellowish-green to brownish fluid over the pylorus (stomach exit) 56 gets into the stomach if it can not prevent reflux. It mixes there with the liquid in which the endoscope capsule floats and tarnishes her. This turbidity can also be detected automatically and thus allows conclusions to be drawn about the functionality of the pylorus. In addition, a liver malfunction can also be detected from the automatically determined color composition of the fluid. Bile occurring in the fluid consists of various chemical components, some of which are clear and others which are colored, so that the color of the bile varies from yellowish to greenish depending on the proportion of the main bile in the bile, namely yellow bilirubin and green biliverdin. Both are breakdown products of the liver of hemoglobin. Depending on the change in the concentrations of one of the two degradation products, the bile fluid thus appears more yellowish or greenish and allows conclusions about liver dysfunctions for certain color shifts. This color difference is retained when mixing with the liquid solution in the working space (stomach and / or intestine) and can be detected automatically by means of an analysis of the endoscope images with methods of digital image processing or analysis.

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• DE 10142253 C1 [0002]

Claims (7)

Method for examining a hollow organ ( 6 ) with a magnet-guided endoscope capsule ( 2 ), from the inside of the hollow organ ( 6 ) continuous video images are generated during the navigation of the endoscope capsule ( 2 ) in the hollow organ ( 6 ) are analyzed by digital image processing methods and automatically analyzed for the presence of diagnostically relevant information, and in which the result of this analysis for controlling the endoscope capsule ( 2 ) is used. The method of claim 1, wherein the diagnostically relevant information is the presence of areas on the wall of the hollow organ that indicate possible gastritis. Method according to Claim 2, in which a gray value image is generated from the video images for localizing high intensity gradients by preprocessing, to each pixel of which a gray value is assigned, from which a filtered image is generated by means of smoothing and edge filtering whose pixels are assigned a filter value, wherein a threshold value is derived from the filter values of the pixels of the filtered image, and in which the video image is divided into a plurality of groups each comprising a plurality of pixels and only those groups are used as relevant groups for analysis in the filtered Image contains a minimum number of pixels that exceed the threshold. The method of claim 3, wherein each relevant group is assigned a plurality of properties, with which it is judged with an expert system trained with a plurality of evaluated test images whether the relevant group indicates a gastric region or not. Method according to Claim 4, in which the test images used for training undergo pre-processing according to Claim 3, and the relevant groups obtained in this way are classified manually as diagnostically relevant or diagnostically irrelevant. The method of claim 1, wherein the diagnostically relevant information consists in a turbidity and / or coloration of a liquid present in the hollow organ and / or a presence of particles in the liquid. Apparatus for inspecting a hollow organ having a magnet system for guiding a permanent-capsule provided with a permanent magnet which receives video images from the interior of the hollow organ and transmits them to a control and evaluation device in which software for carrying out one of the preceding methods is implemented.

Images